debug: force use of newest cloud-build

* updating custom container with PyTorch v1.13

* moved etcd install to custom container build

.cloud-build/execute_changed_notebooks_helper.py

@@ -245,7 +245,7 @@ def process_and_execute_notebook(
         result.logs_bucket = operation_metadata.build.logs_bucket
 
         # Block and wait for the result
-        operation_result = operation.result()
+        operation_result = operation.result(timeout=86400)
 
         result.duration = datetime.datetime.now() - time_start
         result.is_pass = True

community-content/pytorch_efficient_training/README.md

@@ -15,15 +15,19 @@ pip install -r requirements.txt
 * resnet_dp.py - Train ResNet-50 on single node multiple GPUs with `DataParallel` strategy.
 * resnet_ddp.py - Train ResNet-50 on single node multiple GPUs with `DistributedDataParallel` strategy.
 * resnet_ddp_wds.py - Train ResNet-50 on single node multiple GPUs with `DistributedDataParallel` strategy and `Webdataset`.
+* resnet_fsdp.py - Train ResNet-50 on single node multiple GPUs with `FullyShardedDataParallel` strategy.
+* resnet_fsdp_wds.py - Train ResNet-50 on single node multiple GPUs with `FullyShardedDataParallel` strategy and `Webdataset`.
 * shard_imagenet.py - Shard ImagNet individual files into `tar` files.
 
 ## Benchmark
 
 When run the benchmark on Nvidia T4 GPUs using ImageNet validation dataset, you can get the result like:
-Strategy              | Seconds/Epoch - Local Data | Seconds/Epoch - Cloud Data
---------------------- | -------------------------- | --------------------------
-On 1 GPU              | 489                        | 804 (2x slower)
-On 4 GPUs (DP)        | 157                        | 738 (5x slower)
-On 4 GPUs (DDP)       | 134                        | 432 (3x slower)
-On 4 GPUs (DDP + WDS) | 131                        | 133 (same performance)
+Strategy               | Seconds/Epoch - Local Data | Seconds/Epoch - Cloud Data
+---------------------- | -------------------------- | --------------------------
+On 1 GPU               | 489                        | 804 (2x slower)
+On 4 GPUs (DP)         | 157                        | 738 (5x slower)
+On 4 GPUs (DDP)        | 134                        | 432 (3x slower)
+On 4 GPUs (DDP + WDS)  | 131                        | 133 (same performance)
+On 4 GPUs (FSDP)       | 139                        | 353 (3x slower)
+On 4 GPUs (FSDP + WDS) | 138                        | 135 (same performance)
 

community-content/pytorch_efficient_training/resnet_fsdp.py

@@ -0,0 +1,242 @@
+# Copyright 2022 Google LLC
+#
+# Licensed under the Apache License, Version 2.0 (the \"License\");
+# you may not use this file except in compliance with the License.\n",
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an \"AS IS\" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Train resnet on multiple GPUs with FSDP."""
+
+import argparse
+import functools
+import os
+import time
+
+from PIL import Image
+import torch
+from torch import nn
+import torch.distributed as dist
+from torch.distributed.fsdp import FullyShardedDataParallel as FSDP
+from torch.distributed.fsdp.wrap import size_based_auto_wrap_policy
+import torch.multiprocessing as mp
+import torchmetrics
+import torchvision
+from torchvision.models import resnet50
+
+
+class ImageFolder(torchvision.datasets.ImageFolder):
+  """Class for loading imagenet."""
+
+  def __init__(self, image_list_file, transform=None, target_transform=None):
+    self.samples = self._make_dataset(image_list_file)
+    self.loader = self._loader
+
+    self.imgs = self.samples
+    self.targets = [s[1] for s in self.samples]
+
+    self.transform = transform
+    self.target_transform = target_transform
+
+  def _make_dataset(self, image_list_file):
+    items = []
+    with open(image_list_file, 'r') as f:
+      for line in f:
+        item = line.strip().split(' ')
+        items.append((item[0], int(item[1])))
+    return items
+
+  def _loader(self, image_path):
+    with open(image_path, 'rb') as f:
+      img = Image.open(f)
+      img = img.convert('RGB')
+      return img
+
+
+def train(model, device, dataloader, optimizer):
+  model.train()
+  for image, target in dataloader:
+    image = image.to(device, non_blocking=True)
+    target = target.to(device, non_blocking=True)
+    pred = model(image)
+    # pred.shape (N, C), target.shape (N)
+    loss = nn.functional.cross_entropy(pred, target)
+    optimizer.zero_grad()
+    loss.backward()
+    optimizer.step()
+  return loss
+
+
+def evaluate(model, device, dataloader, metric):
+  model.eval()
+  with torch.no_grad():
+    for image, target in dataloader:
+      image = image.to(device, non_blocking=True)
+      target = target.to(device, non_blocking=True)
+      pred = model(image)
+      metric.update(pred, target)
+  accuracy = metric.compute()
+  metric.reset()
+  return accuracy
+
+
+def worker(gpu, args):
+  """Run training and evaluation."""
+  # Init process group.
+  print(f'Initiating process {gpu}')
+  dist.init_process_group(
+      backend='nccl',
+      init_method='env://',
+      world_size=args.gpus,
+      rank=gpu)
+
+  # Create train dataloader.
+  train_dataset = ImageFolder(
+      image_list_file=args.train_data_path,
+      transform=torchvision.transforms.Compose([
+          torchvision.transforms.RandomResizedCrop(224),
+          torchvision.transforms.RandomHorizontalFlip(),
+          torchvision.transforms.ToTensor(),
+          torchvision.transforms.Normalize(
+              mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
+      ]))
+  train_sampler = torch.utils.data.distributed.DistributedSampler(
+      train_dataset, num_replicas=args.gpus, rank=gpu)
+  train_dataloader = torch.utils.data.DataLoader(
+      dataset=train_dataset,
+      batch_size=args.train_batch_size,
+      shuffle=False,
+      num_workers=args.dataloader_num_workers,
+      pin_memory=True,
+      sampler=train_sampler)
+  if gpu == 0:
+    print(f'Train dataloader | samples: {len(train_dataloader.dataset)}, '
+          f'num workers: {train_dataloader.num_workers}, '
+          f'global batch size: {args.train_batch_size * args.gpus}, '
+          f'batches/epoch: {len(train_dataloader)}')
+
+  # Create eval dataloader.
+  eval_dataset = ImageFolder(
+      image_list_file=args.eval_data_path,
+      transform=torchvision.transforms.Compose([
+          torchvision.transforms.Resize(256),
+          torchvision.transforms.CenterCrop(224),
+          torchvision.transforms.ToTensor(),
+          torchvision.transforms.Normalize(
+              mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
+      ]))
+  eval_sampler = torch.utils.data.distributed.DistributedSampler(
+      eval_dataset, num_replicas=args.gpus, rank=gpu)
+  eval_dataloader = torch.utils.data.DataLoader(
+      dataset=eval_dataset,
+      batch_size=args.eval_batch_size,
+      shuffle=False,
+      num_workers=args.dataloader_num_workers,
+      pin_memory=True,
+      drop_last=True,
+      sampler=eval_sampler)
+  if gpu == 0:
+    print(f'Eval dataloader | samples: {len(eval_dataloader.dataset)}, '
+          f'num workers: {eval_dataloader.num_workers}, '
+          f'batch size: {args.eval_batch_size}, '
+          f'batches/epoch: {len(eval_dataloader)}')
+
+  # Wrap policy.
+  my_auto_wrap_policy = functools.partial(
+      size_based_auto_wrap_policy, min_num_params=100)
+  torch.cuda.set_device(gpu)
+
+  # Create model.
+  model = resnet50(weights=None)
+  model.to(args.device)
+  model = FSDP(model, auto_wrap_policy=my_auto_wrap_policy)
+
+  # Optimizer.
+  optimizer = torch.optim.SGD(model.parameters(), 0.1)
+
+  # Main loop.
+  metric = torchmetrics.classification.Accuracy(top_k=1).to(args.device)
+  for epoch in range(1, args.epochs + 1):
+    if gpu == 0:
+      print(f'Running epoch {epoch}')
+    train_sampler.set_epoch(epoch)
+
+    start = time.time()
+    train(model, args.device, train_dataloader, optimizer)
+    end = time.time()
+    if gpu == 0:
+      print(f'Training finished in {(end - start):>0.3f} seconds')
+
+    start = time.time()
+    evaluate(model, args.device, eval_dataloader, metric)
+    end = time.time()
+    if gpu == 0:
+      print(f'Evaluation finished in {(end - start):>0.3f} seconds')
+
+  if gpu == 0:
+    print('Done')
+  dist.destroy_process_group()
+
+
+def create_args():
+  """Create main args."""
+  parser = argparse.ArgumentParser(
+      formatter_class=argparse.ArgumentDefaultsHelpFormatter)
+  parser.add_argument(
+      '--gpus',
+      default=4,
+      type=int,
+      help='number of gpus to use')
+  parser.add_argument(
+      '--epochs',
+      default=2,
+      type=int,
+      help='number of total epochs to run')
+  parser.add_argument(
+      '--dataloader_num_workers',
+      default=2,
+      type=int,
+      help='number of workders for dataloader')
+  parser.add_argument(
+      '--train_data_path',
+      default='',
+      type=str,
+      help='path to training data')
+  parser.add_argument(
+      '--train_batch_size',
+      default=32,
+      type=int,
+      help='batch size for training per gpu')
+  parser.add_argument(
+      '--eval_data_path',
+      default='',
+      type=str,
+      help='path to evaluation data')
+  parser.add_argument(
+      '--eval_batch_size',
+      default=32,
+      type=int,
+      help='batch size for evaluation per gpu')
+  args = parser.parse_args()
+  return args
+
+
+def main():
+  args = create_args()
+
+  os.environ['MASTER_ADDR'] = 'localhost'
+  os.environ['MASTER_PORT'] = '8888'
+
+  args.device = 'cuda' if torch.cuda.is_available() else 'cpu'
+  print(f'Launch job on {args.gpus} GPUs with FSDP')
+  mp.spawn(worker, nprocs=args.gpus, args=(args,))
+
+
+if __name__ == '__main__':
+  main()

community-content/pytorch_efficient_training/resnet_fsdp_wds.py

@@ -0,0 +1,240 @@
+"""Train resnet on multiple GPUs with DDP."""
+
+import argparse
+import functools
+import itertools
+import math
+import os
+import time
+
+import torch
+from torch import nn
+import torch.distributed as dist
+from torch.distributed.fsdp import FullyShardedDataParallel as FSDP
+from torch.distributed.fsdp.wrap import size_based_auto_wrap_policy
+import torch.multiprocessing as mp
+import torchmetrics
+from torchvision.models import resnet50
+from torchvision.transforms import transforms
+import webdataset as wds
+
+
+def wds_split(src, rank, world_size):
+  """Shards split function for webdataset."""
+  # The context of caller of this function is within multiple processes
+  # (by DDP world_size) and multiple workers (by dataloader_num_workers).
+  # So we totally have (world_size * num_workers) workers for processing data.
+  # NOTE: Raw data should be sharded to enough shards to make sure one process
+  # can handle at least one shard, otherwise the process may hang.
+  worker_id = 0
+  num_workers = 1
+  worker_info = torch.utils.data.get_worker_info()
+  if worker_info:
+    worker_id = worker_info.id
+    num_workers = worker_info.num_workers
+  for s in itertools.islice(src, rank * num_workers + worker_id, None,
+                            world_size * num_workers):
+    yield s
+
+
+def identity(x):
+  return x
+
+
+def create_wds_dataloader(rank, args, mode):
+  """Create webdataset dataset and dataloader."""
+  if mode == 'train':
+    transform = transforms.Compose([
+        transforms.RandomResizedCrop(224),
+        transforms.RandomHorizontalFlip(),
+        transforms.ToTensor(),
+        transforms.Normalize(
+            mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
+    ])
+    data_path = args.train_data_path
+    data_size = args.train_data_size
+    batch_size_local = args.train_batch_size
+    batch_size_global = args.train_batch_size * args.gpus
+    # Since webdataset disallows partial batch, we pad the last batch for train.
+    batches = int(math.ceil(data_size / batch_size_global))
+  else:
+    transform = transforms.Compose([
+        transforms.Resize(256),
+        transforms.CenterCrop(224),
+        transforms.ToTensor(),
+        transforms.Normalize(
+            mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
+    ])
+    data_path = args.eval_data_path
+    data_size = args.eval_data_size
+    batch_size_local = args.eval_batch_size
+    batch_size_global = args.eval_batch_size * args.gpus
+    # Since webdataset disallows partial batch, we drop the last batch for eval.
+    batches = int(data_size / batch_size_global)
+
+  dataset = wds.DataPipeline(
+      wds.SimpleShardList(data_path),
+      functools.partial(wds_split, rank=rank, world_size=args.gpus),
+      wds.tarfile_to_samples(),
+      wds.decode('pil'),
+      wds.to_tuple('jpg;png;jpeg cls'),
+      wds.map_tuple(transform, identity),
+      wds.batched(batch_size_local, partial=False),
+  )
+  num_workers = args.dataloader_num_workers
+  dataloader = wds.WebLoader(
+      dataset=dataset,
+      batch_size=None,
+      shuffle=False,
+      num_workers=num_workers,
+      persistent_workers=True if num_workers > 0 else False,
+      pin_memory=True).repeat(nbatches=batches)
+  print(f'{mode} dataloader | samples: {data_size}, '
+        f'num_workers: {num_workers}, '
+        f'local batch size: {batch_size_local}, '
+        f'global batch size: {batch_size_global}, '
+        f'batches: {batches}')
+  return dataloader
+
+
+def train(model, device, dataloader, optimizer):
+  model.train()
+  for image, target in dataloader:
+    image = image.to(device, non_blocking=True)
+    target = target.to(device, non_blocking=True)
+    pred = model(image)
+    # pred.shape (N, C), target.shape (N)
+    loss = nn.functional.cross_entropy(pred, target)
+    optimizer.zero_grad()
+    loss.backward()
+    optimizer.step()
+  return loss
+
+
+def evaluate(model, device, dataloader, metric):
+  model.eval()
+  with torch.no_grad():
+    for image, target in dataloader:
+      image = image.to(device, non_blocking=True)
+      target = target.to(device, non_blocking=True)
+      pred = model(image)
+      metric.update(pred, target)
+  accuracy = metric.compute()
+  metric.reset()
+  return accuracy
+
+
+def worker(gpu, args):
+  """Run training and evaluation."""
+  # Init process group.
+  print(f'Initiating process {gpu}')
+  dist.init_process_group(
+      backend='nccl',
+      init_method='env://',
+      world_size=args.gpus,
+      rank=gpu)
+
+  # Create dataloader.
+  train_dataloader = create_wds_dataloader(gpu, args, 'train')
+  eval_dataloader = create_wds_dataloader(gpu, args, 'eval')
+
+  # Wrap policy.
+  my_auto_wrap_policy = functools.partial(
+      size_based_auto_wrap_policy, min_num_params=100)
+  torch.cuda.set_device(gpu)
+
+  # Create model.
+  model = resnet50(weights=None)
+  model.to(args.device)
+  model = FSDP(model, auto_wrap_policy=my_auto_wrap_policy)
+
+  # Optimizer.
+  optimizer = torch.optim.SGD(model.parameters(), 0.1)
+
+  # Main loop.
+  metric = torchmetrics.classification.Accuracy(top_k=1).to(args.device)
+  for epoch in range(1, args.epochs + 1):
+    if gpu == 0:
+      print(f'Running epoch {epoch}')
+
+    start = time.time()
+    train(model, args.device, train_dataloader, optimizer)
+    end = time.time()
+    if gpu == 0:
+      print(f'Training finished in {(end - start):>0.3f} seconds')
+
+    start = time.time()
+    evaluate(model, args.device, eval_dataloader, metric)
+    end = time.time()
+    if gpu == 0:
+      print(f'Evaluation finished in {(end - start):>0.3f} seconds')
+
+  if gpu == 0:
+    print('Done')
+
+
+def create_args():
+  """Create main args."""
+  parser = argparse.ArgumentParser(
+      formatter_class=argparse.ArgumentDefaultsHelpFormatter)
+  parser.add_argument(
+      '--gpus',
+      default=4,
+      type=int,
+      help='number of gpus to use')
+  parser.add_argument(
+      '--epochs',
+      default=2,
+      type=int,
+      help='number of total epochs to run')
+  parser.add_argument(
+      '--dataloader_num_workers',
+      default=2,
+      type=int,
+      help='number of workders for dataloader')
+  parser.add_argument(
+      '--train_data_path',
+      default='',
+      type=str,
+      help='path to training data')
+  parser.add_argument(
+      '--train_batch_size',
+      default=32,
+      type=int,
+      help='batch size for training per gpu')
+  parser.add_argument(
+      '--train_data_size',
+      default=50000,
+      type=int,
+      help='data size for training')
+  parser.add_argument(
+      '--eval_data_path',
+      default='',
+      type=str,
+      help='path to evaluation data')
+  parser.add_argument(
+      '--eval_batch_size',
+      default=32,
+      type=int,
+      help='batch size for evaluation per gpu')
+  parser.add_argument(
+      '--eval_data_size',
+      default=50000,
+      type=int,
+      help='data size for evaluation')
+  args = parser.parse_args()
+  return args
+
+
+def main():
+  args = create_args()
+  os.environ['MASTER_ADDR'] = 'localhost'
+  os.environ['MASTER_PORT'] = '8888'
+
+  args.device = 'cuda' if torch.cuda.is_available() else 'cpu'
+  print(f'Launch job on {args.gpus} GPUs with FSDP')
+  mp.spawn(worker, nprocs=args.gpus, args=(args,))
+
+
+if __name__ == '__main__':
+  main()

notebooks/community/ml_ops/stage5/get_started_with_vertex_endpoint_and_shared_vm.ipynb

@@ -159,9 +159,9 @@
         "\n",
         "# Install the packages\n",
         "\n",
-        "! pip3 install --upgrade google-cloud-aiplatform $USER_FLAG -q\n",
-        "! pip3 install --upgrade tensorflow $USER_FLAG -q\n",
-        "! pip3 install --upgrade tensorflow-hub $USER_FLAG -q"
+        "! pip3 install --upgrade google-cloud-aiplatform \\\n",
+        "                         tensorflow \\\n",
+        "                         tensorflow-hub $USER_FLAG -q"
       ]
     },
     {
@@ -307,22 +307,29 @@
         "id": "timestamp"
       },
       "source": [
-        "#### Timestamp\n",
+        "#### UUID\n",
         "\n",
-        "If you are in a live tutorial session, you might be using a shared test account or project. To avoid name collisions between users on resources created, you create a timestamp for each instance session, and append the timestamp onto the name of resources you create in this tutorial."
+        "If you are in a live tutorial session, you might be using a shared test account or project. To avoid name collisions between users on resources created, you create a uuid for each instance session, and append it onto the name of resources you create in this tutorial."
       ]
     },
     {
       "cell_type": "code",
       "execution_count": null,
       "metadata": {
-        "id": "timestamp"
+        "id": "84Vdv7R-QEH6"
       },
       "outputs": [],
       "source": [
-        "from datetime import datetime\n",
+        "import random\n",
+        "import string\n",
         "\n",
-        "TIMESTAMP = datetime.now().strftime(\"%Y%m%d%H%M%S\")"
+        "\n",
+        "# Generate a uuid of a specifed length(default=8)\n",
+        "def generate_uuid(length: int = 8) -> str:\n",
+        "    return \"\".join(random.choices(string.ascii_lowercase + string.digits, k=length))\n",
+        "\n",
+        "\n",
+        "UUID = generate_uuid()"
       ]
     },
     {
@@ -421,7 +428,7 @@
       "outputs": [],
       "source": [
         "if BUCKET_URI == \"\" or BUCKET_URI is None or BUCKET_URI == \"gs://[your-bucket-name]\":\n",
-        "    BUCKET_NAME = PROJECT_ID + \"aip-\" + TIMESTAMP\n",
+        "    BUCKET_NAME = PROJECT_ID + \"aip-\" + UUID\n",
         "    BUCKET_URI = \"gs://\" + BUCKET_NAME"
       ]
     },
@@ -523,7 +530,7 @@
         "\n",
         "Setup up the following constants for Vertex AI:\n",
         "\n",
-        "- `API_ENDPOINT`: The Vertex AI API service endpoint for `Endpoint` services."
+        "- `API_ENDPOINT`: The Vertex AI API service endpoint."
       ]
     },
     {
@@ -538,46 +545,10 @@
         "API_ENDPOINT = \"{}-aiplatform.googleapis.com\".format(REGION)\n",
         "\n",
         "# Vertex location root path for your dataset, model and endpoint resources\n",
-        "PARENT = \"projects/\" + PROJECT_ID + \"/locations/\" + REGION"
-      ]
-    },
-    {
-      "cell_type": "markdown",
-      "metadata": {
-        "id": "clients:metadata"
-      },
-      "source": [
-        "## Set up clients\n",
+        "PARENT = \"projects/\" + PROJECT_ID + \"/locations/\" + REGION\n",
         "\n",
-        "The Vertex  works as a client/server model. On your side (the Python script) you will create a client that sends requests and receives responses from the Vertex AI server.\n",
-        "\n",
-        "You will use different clients in this tutorial for different steps in the workflow. So set them all up upfront.\n",
-        "\n",
-        "- Endpoint Service for creating endpoints, and deploying models to endpoints."
-      ]
-    },
-    {
-      "cell_type": "code",
-      "execution_count": null,
-      "metadata": {
-        "id": "clients:metadata"
-      },
-      "outputs": [],
-      "source": [
         "# client options same for all services\n",
-        "client_options = {\"api_endpoint\": API_ENDPOINT}\n",
-        "\n",
-        "\n",
-        "def create_endpoint_client():\n",
-        "    client = aip_beta.EndpointServiceClient(client_options=client_options)\n",
-        "    return client\n",
-        "\n",
-        "\n",
-        "clients = {}\n",
-        "clients[\"endpoint\"] = create_endpoint_client()\n",
-        "\n",
-        "for client in clients.items():\n",
-        "    print(client)"
+        "client_options = {\"api_endpoint\": API_ENDPOINT}"
       ]
     },
     {
@@ -592,7 +563,7 @@
         "\n",
         "Set the variables `DEPLOY_GPU/DEPLOY_NGPU` to use a container image supporting a GPU and the number of GPUs allocated to the virtual machine (VM) instance. For example, to use a GPU container image with 4 Nvidia Telsa K80 GPUs allocated to each VM, you would specify:\n",
         "\n",
-        "    (aip.AcceleratorType.NVIDIA_TESLA_K80, 4)\n",
+        "    (aip.gapic.AcceleratorType.NVIDIA_TESLA_K80, 4)\n",
         "\n",
         "\n",
         "Otherwise specify `(None, None)` to use a container image to run on a CPU.\n",
@@ -902,7 +873,7 @@
       "outputs": [],
       "source": [
         "model_icn = aiplatform.Model.upload(\n",
-        "    display_name=\"icn_\" + TIMESTAMP,\n",
+        "    display_name=\"icn_\" + UUID,\n",
         "    artifact_uri=MODEL_ICN_DIR,\n",
         "    serving_container_image_uri=DEPLOY_IMAGE,\n",
         ")\n",
@@ -1013,7 +984,7 @@
       "outputs": [],
       "source": [
         "model_use = aiplatform.Model.upload(\n",
-        "    display_name=\"icn_\" + TIMESTAMP,\n",
+        "    display_name=\"icn_\" + UUID,\n",
         "    artifact_uri=MODEL_USE_DIR,\n",
         "    serving_container_image_uri=DEPLOY_IMAGE,\n",
         ")\n",
@@ -1029,64 +1000,52 @@
       "source": [
         "## Creating a deployment resource pool\n",
         "\n",
-        "Currently, creating deploynent resource pools is only supported via the REST-based API (e.g., CURL).\n",
+        "Currently, creating deploynent resource pools is only supported via the REST-based API (e.g., CURL) and GAPIC APIs (Python).\n",
         "\n",
-        "Use `CreateDeploymentResourcePool` API to create a resource pool, with the following configuration:\n",
+        "Use `create_deployment_resource_pool` API to create a resource pool, with the following configuration:\n",
         "\n",
         "- `dedicated_resources`: Compute (HW) resources to allocate for the shared vm.\n",
         "- `min_replica_count`: Auto-scaling, the minimum number of compute nodes.\n",
         "- `max_replica_count`: Auto-scaling, the maximum number of compute nodes.\n",
         "\n",
-        "Learn more about [Deployment Resource Pools]()."
+        "Learn more about [Deployment Resource Pools](https://googleapis.dev/python/aiplatform/latest/aiplatform_v1beta1/deployment_resource_pool_service.html)."
       ]
     },
     {
       "cell_type": "code",
       "execution_count": null,
       "metadata": {
-        "id": "YiBmoiWYcMQt"
+        "id": "90c51b6cf34a"
       },
       "outputs": [],
       "source": [
-        "DEPLOYMENT_RESOURCE_POOL_ID = \"shared-vm\"  # @param {type: \"string\"}"
-      ]
-    },
-    {
-      "cell_type": "code",
-      "execution_count": null,
-      "metadata": {
-        "id": "0CHPJ4h-Slgs"
-      },
-      "outputs": [],
-      "source": [
-        "import json\n",
-        "import pprint\n",
-        "pp = pprint.PrettyPrinter(indent=4)\n",
-        "\n",
+        "DEPLOYMENT_RESOURCE_POOL_ID = f\"shared-vm-{UUID}\"  # @param {type: \"string\"}\n",
         "MIN_NODES = 1\n",
         "MAX_NODES = 2\n",
         "\n",
-        "CREATE_RP_PAYLOAD = {\n",
-        "  \"deployment_resource_pool\":{\n",
-        "    \"dedicated_resources\":{\n",
-        "      \"machine_spec\":{\n",
-        "        \"machine_type\": DEPLOY_COMPUTE\n",
-        "      },\n",
-        "      \"min_replica_count\": MIN_NODES, \n",
-        "      \"max_replica_count\": MAX_NODES\n",
-        "    }\n",
-        "  },\n",
-        "  \"deployment_resource_pool_id\":DEPLOYMENT_RESOURCE_POOL_ID\n",
-        "}\n",
-        "CREATE_RP_REQUEST=json.dumps(CREATE_RP_PAYLOAD)\n",
-        "pp.pprint(\"CREATE_RP_REQUEST: \" + CREATE_RP_REQUEST)\n",
+        "# Initialize request argument(s)\n",
+        "deployment_resource_pool = aip_beta.DeploymentResourcePool()\n",
+        "deployment_resource_pool.dedicated_resources.min_replica_count = MIN_NODES\n",
+        "deployment_resource_pool.dedicated_resources.max_replica_count = MAX_NODES\n",
+        "deployment_resource_pool.dedicated_resources.machine_spec.machine_type = DEPLOY_COMPUTE\n",
         "\n",
-        "! curl \\\n",
-        "-X POST \\\n",
-        "-H \"Authorization: Bearer $(gcloud auth print-access-token)\" \\\n",
-        "-H \"Content-Type: application/json\" \\\n",
-        "https://{REGION}-aiplatform.googleapis.com/v1beta1/projects/{PROJECT_ID}/locations/{REGION}/deploymentResourcePools \\\n",
-        "-d '{CREATE_RP_REQUEST}'"
+        "request = aip_beta.CreateDeploymentResourcePoolRequest(\n",
+        "    parent=f\"projects/{PROJECT_ID}/locations/{REGION}\",\n",
+        "    deployment_resource_pool=deployment_resource_pool,\n",
+        "    deployment_resource_pool_id=DEPLOYMENT_RESOURCE_POOL_ID,\n",
+        ")\n",
+        "\n",
+        "pool_client = aip_beta.services.deployment_resource_pool_service.DeploymentResourcePoolServiceClient(\n",
+        "    client_options=client_options\n",
+        ")\n",
+        "\n",
+        "op = pool_client.create_deployment_resource_pool(request=request)\n",
+        "print(op)\n",
+        "\n",
+        "result = op.result()\n",
+        "print(result)\n",
+        "\n",
+        "deployment_pool_id = result.name"
       ]
     },
     {
@@ -1099,21 +1058,19 @@
         "\n",
         "Use `GetDeploymentResourcePool` API to check out the deploynent resource pool that you created. \n",
         "\n",
-        "Learn more about [Get Deployment Resource Pool](https://source.corp.google.com/piper///depot/google3/google/cloud/aiplatform/master/deployment_resource_pool_service.proto;l=75?q=deployment_resource_pool&sq=package:piper%20file:%2F%2Fdepot%2Fgoogle3%20-file:google3%2Fexperimental)."
+        "Learn more about [Get Deployment Resource Pool](https://googleapis.dev/python/aiplatform/latest/aiplatform_v1beta1/deployment_resource_pool_service.html)."
       ]
     },
     {
       "cell_type": "code",
       "execution_count": null,
       "metadata": {
-        "id": "6wTLyhPraFah"
+        "id": "b740253903c0"
       },
       "outputs": [],
       "source": [
-        "! curl -X GET \\\n",
-        "-H \"Authorization: Bearer $(gcloud auth print-access-token)\" \\\n",
-        "-H \"Content-Type: application/json\" \\\n",
-        "https://{REGION}-aiplatform.googleapis.com/v1beta1/projects/{PROJECT_ID}/locations/{REGION}/deploymentResourcePools/{DEPLOYMENT_RESOURCE_POOL_ID}"
+        "response = pool_client.get_deployment_resource_pool(name=deployment_pool_id)\n",
+        "print(response)"
       ]
     },
     {
@@ -1126,21 +1083,22 @@
         "\n",
         "Use `ListDeploymentResourcePools` API to list all the deployment resource pools. \n",
         "\n",
-        "Learn more about [Listing Deployment Resource Pools](https://source.corp.google.com/piper///depot/google3/google/cloud/aiplatform/master/deployment_resource_pool_service.proto;l=101?q=deployment_resource_pool&sq=package:piper%20file:%2F%2Fdepot%2Fgoogle3%20-file:google3%2Fexperimental)."
+        "Learn more about [Listing Deployment Resource Pools](https://googleapis.dev/python/aiplatform/latest/aiplatform_v1beta1/deployment_resource_pool_service.html)."
       ]
     },
     {
       "cell_type": "code",
       "execution_count": null,
       "metadata": {
-        "id": "Pxls4sNnaltU"
+        "id": "3ebfd007bff2"
       },
       "outputs": [],
       "source": [
-        "! curl -X GET \\\n",
-        "-H \"Authorization: Bearer $(gcloud auth print-access-token)\" \\\n",
-        "-H \"Content-Type: application/json\" \\\n",
-        "https://{REGION}-aiplatform.googleapis.com/v1beta1/projects/{PROJECT_ID}/locations/{REGION}/deploymentResourcePools"
+        "pools = pool_client.list_deployment_resource_pools(\n",
+        "    parent=f\"projects/{PROJECT_ID}/locations/{REGION}\"\n",
+        ")\n",
+        "for pool in pools:\n",
+        "    print(pool)"
       ]
     },
     {
@@ -1170,11 +1128,11 @@
       },
       "outputs": [],
       "source": [
-        "endpoint_icn = aiplatform.Endpoint.create(display_name=\"icn_\" + TIMESTAMP)\n",
+        "endpoint_icn = aiplatform.Endpoint.create(display_name=\"icn_\" + UUID)\n",
         "\n",
         "print(endpoint_icn)\n",
         "\n",
-        "endpoint_use = aiplatform.Endpoint.create(display_name=\"use_\" + TIMESTAMP)\n",
+        "endpoint_use = aiplatform.Endpoint.create(display_name=\"use_\" + UUID)\n",
         "\n",
         "print(endpoint_use)"
       ]
@@ -1204,6 +1162,12 @@
       },
       "outputs": [],
       "source": [
+        "import json\n",
+        "import pprint\n",
+        "\n",
+        "pp = pprint.PrettyPrinter(indent=4)\n",
+        "\n",
+        "\n",
         "SHARED_RESOURCE = \"projects/{project_id}/locations/{region}/deploymentResourcePools/{deployment_resource_pool_id}\".format(\n",
         "    project_id=PROJECT_ID,\n",
         "    region=REGION,\n",
@@ -1363,18 +1327,27 @@
         "    time.sleep(30)"
       ]
     },
+    {
+      "cell_type": "markdown",
+      "metadata": {
+        "id": "52248c450776"
+      },
+      "source": [
+        "### Get deployment details for the endpoint\n",
+        "\n",
+        "List the deployed models on the endpoint."
+      ]
+    },
     {
       "cell_type": "code",
       "execution_count": null,
       "metadata": {
-        "id": "86a659bf60f0"
+        "id": "3b768614e7c6"
       },
       "outputs": [],
       "source": [
-        "! curl -X GET \\\n",
-        " -H \"Authorization: Bearer $(gcloud auth print-access-token)\" \\\n",
-        " -H \"Content-Type: application/json\" \\\n",
-        "https://{REGION}-aiplatform.googleapis.com/v1/projects/759209241365/locations/us-central1/endpoints/2259566763823857664"
+        "print(endpoint_icn.list_models())\n",
+        "print(endpoint_use.list_models())"
       ]
     },
     {
@@ -1557,21 +1530,19 @@
       "source": [
         "#### Delete the `DeploymentResourcePool`\n",
         "\n",
-        "The method 'delete()' will delete your deployment resource pool."
+        "The method 'delete_deployment_resource_pool()' will delete your deployment resource pool."
       ]
     },
     {
       "cell_type": "code",
       "execution_count": null,
       "metadata": {
-        "id": "ac40cc1d594a"
+        "id": "b76a4de1e57e"
       },
       "outputs": [],
       "source": [
-        "! curl -X DELETE \\\n",
-        "-H \"Authorization: Bearer $(gcloud auth print-access-token)\" \\\n",
-        "-H \"Content-Type: application/json\" \\\n",
-        "https://{REGION}-aiplatform.googleapis.com/v1beta1/projects/{PROJECT_ID}/locations/{REGION}/deploymentResourcePools/{DEPLOYMENT_RESOURCE_POOL_ID}"
+        "response = pool_client.delete_deployment_resource_pool(name=deployment_pool_id)\n",
+        "print(response)"
       ]
     },
     {

notebooks/community/sdk/sdk_pytorch_torchrun_custom_container_training_imagenet.ipynb

@@ -567,7 +567,7 @@
       "outputs": [],
       "source": [
         "%%writefile trainer/Dockerfile\n",
-        "FROM gcr.io/deeplearning-platform-release/pytorch-gpu.1-12\n",
+        "FROM gcr.io/deeplearning-platform-release/pytorch-gpu.1-13:m102\n",
         "\n",
         "RUN curl https://packages.cloud.google.com/apt/doc/apt-key.gpg | apt-key add - && \\\n",
         "    # Install reduction server plugin on GPU containers. google-fast-socket is\n",
@@ -588,16 +588,31 @@
         "RUN apt-get update -y && \\\n",
         "    apt-get install -y curl gnupg telnet nano net-tools iputils-ping\n",
         "\n",
+        "# Set ETCD version\n",
+        "ARG ETCD_VER=v2.3.0\n",
+        "# Choose either URL\n",
+        "ARG GOOGLE_URL=https://storage.googleapis.com/etcd\n",
+        "ARG GITHUB_URL=https://github.com/etcd-io/etcd/releases/download\n",
+        "# Set ETCD URL to download from\n",
+        "ARG DOWNLOAD_URL=$GOOGLE_URL\n",
         "\n",
+        "# Install ETCD\n",
+        "RUN mkdir -p /tmp/etcd-download-test && \\\n",
+        "    curl -L ${DOWNLOAD_URL}/${ETCD_VER}/etcd-${ETCD_VER}-linux-amd64.tar.gz -o /tmp/etcd-${ETCD_VER}-linux-amd64.tar.gz && \\\n",
+        "    tar xzvf /tmp/etcd-${ETCD_VER}-linux-amd64.tar.gz -C /tmp/etcd-download-test --strip-components=1 && \\\n",
+        "    rm -f /tmp/etcd-${ETCD_VER}-linux-amd64.tar.gz\n",
+        "\n",
+        "# Copy training application code\n",
         "COPY . /trainer\n",
         "\n",
         "WORKDIR /trainer\n",
         "\n",
+        "# Install dependencies\n",
         "RUN pip install -r requirements.txt\n",
         "\n",
         "RUN chmod 777 main.sh\n",
         "\n",
-        "# download data to the container\n",
+        "# Download data to the container\n",
         "RUN wget -q -P /trainer/data https://image-net.org/data/tiny-imagenet-200.zip\n",
         "RUN unzip -q /trainer/data/tiny-imagenet-200.zip\n",
         "RUN rm /trainer/data/tiny-imagenet-200.zip\n",
@@ -614,8 +629,8 @@
       "outputs": [],
       "source": [
         "%%writefile trainer/requirements.txt\n",
-        "torch==1.12.0\n",
-        "torchvision==0.13.0\n",
+        "torch==1.13.0\n",
+        "torchvision==0.14.0\n",
         "tensorboard==2.5.0\n",
         "protobuf==3.20.*\n",
         "python-etcd\n",
@@ -675,30 +690,17 @@
         "setup_etcd() {\n",
         "    HOST_IP=$1\n",
         "    # Start a local instane of ETCD v2 \n",
-        "    ETCD_VER=v2.3.0 #v3.5.6\n",
         "    export ETCD_ENABLE_V2=true\n",
         "    export ETCDCTL_API=2\n",
         "\n",
-        "    # choose either URL\n",
-        "    GOOGLE_URL=https://storage.googleapis.com/etcd\n",
-        "    GITHUB_URL=https://github.com/etcd-io/etcd/releases/download\n",
-        "    DOWNLOAD_URL=${GOOGLE_URL}\n",
-        "\n",
-        "    rm -f /tmp/etcd-${ETCD_VER}-linux-amd64.tar.gz\n",
-        "    rm -rf /tmp/etcd-download-test && mkdir -p /tmp/etcd-download-test\n",
-        "\n",
-        "    curl -L ${DOWNLOAD_URL}/${ETCD_VER}/etcd-${ETCD_VER}-linux-amd64.tar.gz -o /tmp/etcd-${ETCD_VER}-linux-amd64.tar.gz\n",
-        "    tar xzvf /tmp/etcd-${ETCD_VER}-linux-amd64.tar.gz -C /tmp/etcd-download-test --strip-components=1\n",
-        "    rm -f /tmp/etcd-${ETCD_VER}-linux-amd64.tar.gz\n",
-        "\n",
         "    /tmp/etcd-download-test/etcd --name s1 --data-dir /tmp/etcd-download-test/s1  \\\n",
         "    --listen-client-urls http://0.0.0.0:2379 --advertise-client-urls http://$HOST_IP:2379 \\\n",
         "    --listen-peer-urls http://0.0.0.0:2380 --initial-advertise-peer-urls http://$HOST_IP:2380 \\\n",
         "    --initial-cluster s1=http://$HOST_IP:2380 --initial-cluster-token tkn \\\n",
         "    --initial-cluster-state new &> /tmp/etcd-download-test/node.log &\n",
         "\n",
-        "    sudo /tmp/etcd-download-test/etcd --version\n",
-        "    sudo /tmp/etcd-download-test/etcdctl --version\n",
+        "    /tmp/etcd-download-test/etcd --version\n",
+        "    /tmp/etcd-download-test/etcdctl --version\n",
         "}\n",
         "\n",
         "\n",

notebooks/official/tabular_workflows/tabnet_on_vertex_pipelines.ipynb

@@ -97,7 +97,7 @@
         "### Dataset\n",
         "\n",
         "The dataset you will be using is [Bank Marketing](https://archive.ics.uci.edu/ml/datasets/bank+marketing).\n",
-        "The data is for direct marketing campaigns (phone calls) of a Portuguese banking institution. The binary classification goal is to predict if a client subscribe a term deposit. For this notebook, you randomly selected 90% of the rows in the original dataset and saved them in a train.csv file hosted on Cloud Storage. To download the file, click [here](https://storage.googleapis.com/cloud-samples-data/vertex-ai/tabular-workflows/datasets/bank-marketing/train.csv)."
+        "The data is for direct marketing campaigns (phone calls) of a Portuguese banking institution. The binary classification goal is to predict if a client subscribe a term deposit. For this notebook, you randomly selected 90% of the rows in the original dataset and saved them in a train.csv file hosted on Cloud Storage. To download the file, click [here](https://storage.googleapis.com/cloud-samples-data-us-central1/vertex-ai/tabular-workflows/datasets/bank-marketing/train.csv)."
       ]
     },
     {
@@ -746,7 +746,7 @@
       },
       "outputs": [],
       "source": [
-        "data_source_csv_filenames = \"gs://cloud-samples-data/vertex-ai/tabular-workflows/datasets/bank-marketing/train.csv\"\n",
+        "data_source_csv_filenames = \"gs://cloud-samples-data-us-central1/vertex-ai/tabular-workflows/datasets/bank-marketing/train.csv\"\n",
         "data_source_bigquery_table_path = (\n",
         "    None  # @param {type:\"string\"}, format: bq://bq_project.bq_dataset.bq_table\n",
         ")"

notebooks/official/tabular_workflows/wide_and_deep_on_vertex_pipelines.ipynb

@@ -97,7 +97,7 @@
         "### Dataset\n",
         "\n",
         "The dataset you will be using is [Bank Marketing](https://archive.ics.uci.edu/ml/datasets/bank+marketing).\n",
-        "The data is for direct marketing campaigns (phone calls) of a Portuguese banking institution. The binary classification goal is to predict if a client will subscribe a term deposit. For this notebook, we randomly selected 90% of the rows in the original dataset and saved them in a train.csv file hosted on Cloud Storage. To download the file, click [here](https://storage.googleapis.com/cloud-samples-data/vertex-ai/tabular-workflows/datasets/bank-marketing/train.csv)."
+        "The data is for direct marketing campaigns (phone calls) of a Portuguese banking institution. The binary classification goal is to predict if a client will subscribe a term deposit. For this notebook, we randomly selected 90% of the rows in the original dataset and saved them in a train.csv file hosted on Cloud Storage. To download the file, click [here](https://storage.googleapis.com/cloud-samples-data-us-central1/vertex-ai/tabular-workflows/datasets/bank-marketing/train.csv)."
       ]
     },
     {
@@ -667,7 +667,7 @@
       },
       "outputs": [],
       "source": [
-        "data_source_csv_filenames = \"gs://cloud-samples-data/vertex-ai/tabular-workflows/datasets/bank-marketing/train.csv\"\n",
+        "data_source_csv_filenames = \"gs://cloud-samples-data-us-central1/vertex-ai/tabular-workflows/datasets/bank-marketing/train.csv\"\n",
         "data_source_bigquery_table_path = (\n",
         "    None  # @param {type:\"string\"}, format: bq://bq_project.bq_dataset.bq_table\n",
         ")"

notebooks/official/tensorboard/tensorboard_profiler_custom_training.ipynb

@@ -29,7 +29,7 @@
         "id": "l2mMvIUG9meX"
       },
       "source": [
-        "# Profile model training performance using Profiler\n",
+        "# Profile model training performance using Vertex AI TensorBoard Profiler\n",
         "\n",
         "<table align=\"left\">\n",
         "\n",
@@ -49,7 +49,7 @@
         "      <img src=\"https://lh3.googleusercontent.com/UiNooY4LUgW_oTvpsNhPpQzsstV5W8F7rYgxgGBD85cWJoLmrOzhVs_ksK_vgx40SHs7jCqkTkCk=e14-rj-sc0xffffff-h130-w32\" alt=\"Vertex AI logo\">\n",
         "      Open in Vertex AI Workbench\n",
         "    </a>\n",
-        "  </td>                                                                                               \n",
+        "  </td>\n",
         "</table>"
       ]
     },
@@ -331,22 +331,16 @@
         "! gcloud projects add-iam-policy-binding $PROJECT_ID \\\n",
         "        --member=\"serviceAccount:$SERVICE_ACCOUNT\" \\\n",
         "        --role=\"roles/storage.admin\" \\\n",
-        "        --quiet"
-      ]
-    },
-    {
-      "cell_type": "code",
-      "execution_count": null,
-      "metadata": {
-        "id": "S_8_5jm-Gk6w"
-      },
-      "outputs": [],
-      "source": [
+        "        --quiet\n",
+        "\n",
         "# Grant AI Platform permission.\n",
         "! gcloud projects add-iam-policy-binding $PROJECT_ID \\\n",
         "        --member=\"serviceAccount:$SERVICE_ACCOUNT\" \\\n",
         "        --role=\"roles/aiplatform.user\" \\\n",
-        "        --quiet"
+        "        --quiet\n",
+        "\n",
+        "! gcloud projects get-iam-policy $PROJECT_ID \\\n",
+        "        --filter=bindings.members:serviceAccount:$SERVICE_ACCOUNT"
       ]
     },
     {
@@ -768,7 +762,7 @@
         "WORKDIR /root\n",
         "\n",
         "# Installs additional packages as you need.\n",
-        "RUN pip3 install google-cloud-aiplatform[cloud_profiler]\n",
+        "RUN pip3 install google-cloud-aiplatform[cloud_profiler]>=1.19.1\n",
         "\n",
         "# Copies the trainer code to the docker image.\n",
         "RUN mkdir /root/trainer\n",
@@ -798,7 +792,7 @@
         "IMAGE_NAME = \"tensorboard-custom-container\"\n",
         "IMAGE_URI = f\"{REGION}-docker.pkg.dev/{PROJECT_ID}/{DOCKER_REPOSITORY}/{IMAGE_NAME}\"\n",
         "\n",
-        "! gcloud builds submit --project {PROJECT_ID} --region={REGION} --tag {IMAGE_URI} --timeout=60m --quiet"
+        "! gcloud builds submit --project {PROJECT_ID} --region={REGION} --tag {IMAGE_URI} --timeout=3600s --quiet"
       ]
     },
     {